hpc: PBS job scripts for Imperial CX3 (probe + scaled LLM merge run)

Enable running the LLM tier on Imperial's HPC (scheduler: PBS Pro / qsub).
- hpc/probe.pbs: 10-min 1-GPU reconnaissance job resolving the two unknowns
  the RCS docs omit -- compute-node internet access and the L40S driver's
  CUDA version -- plus TMPDIR/disk and available python/cuda modules.
- hpc/llm_merge.pbs: scaled run on an L40S (48 GB), offline HF-cache wired,
  runs configs/llm/merge_hpc.yaml.
- configs/llm/merge_hpc.yaml: Qwen2.5-7B-Instruct (fits the L40S) to reduce
  the noise that left the 0.5B prototype's overall-exceeds sign marginal.
- hpc/README.md: the git-based workflow (login-node uv env + model
  pre-download -> qsub -> rsync results back), confirmed PBS/GPU directives,
  and the code TODOs for the definitive run (multi-seed, more families,
  directed/dilution-resistant merge).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

configs/llm/merge_hpc.yaml

@@ -0,0 +1,20 @@
+experiment: llm_merge_hpc
+kind: llm_merge
+seed: 1
+n_replicates: 1
+
+# Scaled version of configs/llm/merge.yaml for an L40S (48 GB): a bigger, more capable base so the
+# specialists decorrelate cleanly and the "recombined model exceeds any single specialist" sign is
+# less noisy than the 0.5B local prototype (where it was only marginal on overall accuracy).
+# NOTE (code TODOs for the *definitive* run, not yet built): loop several seeds and report mean±CI;
+# add more task families; and add a dilution-resistant / offspring-selected ("directed sex") merge.
+
+base_model: Qwen/Qwen2.5-7B-Instruct    # ~15 GB bf16 + LoRA fits the L40S; fallback: 3B-Instruct
+families: [lists, strings, arith]
+n_train: 800
+n_test: 200
+epochs: 3
+lora: {r: 16, alpha: 32}
+merges: [soup, ties]
+
+output: {dir: results/llm_merge_hpc}